Re: cauchy.m in optim package?

[Juan Pablo Carbajal]

On Fri, Jul 25, 2014 at 10:16 AM, Olaf Till <address@hidden> wrote:
> On Thu, Jul 24, 2014 at 10:10:39PM +0200, Juan Pablo Carbajal wrote:
>> <snip>
>> Do you have a suggestion where to place it?
>
> Some more tests show that a relyable first derivative at some `x'
> seems available if `N' is set to `2' and `r' to a small value. But
> still I see no advantage (for optim) even in the scalar case to using
> the complex step derivative we already have (jacobs.m from Fotios
> Kasolis). And for nonscalar gradients, using something similar to
> `cauchy.m' would mean to use `fft' multiple times for each gradient(?)
> --- disadvantage. And I see no use in optimization for higher order
> derivatives with respect to only one parameter. So I don't see a use
> in `optim'.
>
> Though it's not my field, I don't doubt that the Taylor expansion in
> `cauchy' is useful, and there was a comment in the thread you
> mentioned seeming to say the same. But I see no fitting package. So
> probably `general' or `miscellaneous', though I don't like this
> suggestion.
>
> Olaf
>
> --
> public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

Can be moved to general... if nobody is against that.